SpeaklySpeakly

Other Providers

ElevenLabs, Mistral, Deepgram, Google Gemini and custom endpoints

Besides OpenAI and Groq, the Your API Key engine supports ElevenLabs, Mistral, Deepgram, Google (Gemini 3.5 Transcribe) and any OpenAI-compatible endpoint of your own.

They are all configured in the same place: Settings → Transcription → engine Your API Key → pick the provider in the Provider row → paste the key → Test Connection. The app's own summary applies to every one of them: the key is encrypted and stored locally, calls go directly from Speakly to the provider rather than through Speakly's servers, the provider bills you for usage, and an internet connection is required.

ElevenLabs

  1. Create a key at elevenlabs.io/app/developers/api-keys
  2. Paste it into ElevenLabs API Key — it starts with sk_
  3. Click Test Connection and wait for "API key is valid"

There is no model dropdown: the provider card is labelled Scribe v1 and the panel states "Model: Scribe v1 - ElevenLabs' speech-to-text model with support for 99 languages, speaker diarization, and audio event detection." ElevenLabs and Google (Gemini 3.5 Transcribe Live) give Live Transcription a true realtime stream; every other engine uses the chunked path.

Mistral (Voxtral)

  1. Create a key at console.mistral.ai/api-keys
  2. Paste it into Mistral API Key
  3. Click Test Connection
  4. Pick a model under Model — the list is fetched from your Mistral account once the key is valid; the default is voxtral-mini-latest

Deepgram

  1. Create a key at console.deepgram.com/settings/api-keys
  2. Paste it into Deepgram API Key
  3. Click Test Connection
  4. Pick a model under Model — fetched from your account and trimmed to the newest Nova generation (general and medical); the rest of Deepgram's catalogue is older, tuned for English call audio, or not speech-to-text at all. The stored default is nova-3

The panel notes "Free tier includes $200 credit."

Google — Gemini 3.5 Transcribe

You need a Google AI Studio key
The Google engine runs on the Gemini API with the Gemini 3.5 Transcribe model, so the key comes from Google AI Studio. A Google Cloud Console key made for Speech-to-Text is rejected — the panel answers "This key cannot call the Gemini API" — replace it and click Test Connection. The key is also saved for Google Gemini post-processing, so you only paste it once.
  1. Open aistudio.google.com/api-keys
  2. Click Create API key — it starts with AIza
  3. Paste it into Google AI Studio API Key and click Test Connection
  4. Choose Smart or Verbatim under Transcription Mode

There is one model, so there is no model dropdown. Smart (Recommended) returns clean, readable text — filler words, repetitions and false starts removed, punctuation tidied; Verbatim keeps exactly what was said. The mode applies to dictation, file and YouTube transcription and Live Transcription alike. The language is detected automatically unless you set a spoken language, and your Dictionary is sent as custom vocabulary (up to 100 entries). Pricing is about $0.005 per minute of audio (about $0.009 for the live stream); A key on a project with no billing account gets a rate-limited free tier; once billing is linked (new accounts prepay, minimum $5) every call is billed.

Custom OpenAI-compatible endpoint

The Custom provider points Speakly at any service that implements OpenAI's transcription API — the panel gives the examples: "Connect to any service that implements the OpenAI Whisper API format, such as LocalAI, vLLM, or your own hosted endpoint."

  1. Pick Custom in the Provider row
  2. Base URL: the API root, e.g. https://api.example.com/v1
  3. API Key: the token your endpoint expects
  4. Model Name: the model id to request — whisper-1 by default, "e.g., whisper-1, whisper-large-v3"
  5. Click Test Connection; "Configuration valid" means all three fields are saved
  6. Clear wipes the base URL, key and model name
Other Providers — Speakly